一种基于支持向量机的含噪语音的清/浊/静音分类的新方法  

A Method for Voiced/Unvoiced/Silence Classification of Speech with Noise Using SVM

作　　者：齐峰岩[1] 鲍长春[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100022

出　　处：《电子学报》2006年第4期605-611,共7页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60372063);北京市自然科学基金(No.4042009);北京市教委科技发展项目(KM200310005024)

摘　　要：本文将支持向量机(SVM)方法应用于语音信号的清/浊/静音检测中,提出并验证了一种在各种信噪比等级下将语音信号有效地分为清音、浊音和静音三类信号的新型分类算法.首先,在高信噪比情况下,本文采用了G.729B VAD中的四个差分参数作为SVM分类器的输入特征参数,进行了静音分类的对比实验,得到了优于G.729B VAD和BP神经网络传统算法的实验结果,说明引入这种机器学习方法做语音分类是可行的,并分析讨论了在核函数不同的情况下支持向量机在实验中所表现出的性能.其次,又讨论了在低信噪比条件下,如何通过对含噪语音建立统计模型,提取对噪音免疫的统计特征参数,并给出了一种对时变背景噪声自适应的估计方法.最后,通过在不同噪音环境下的对比实验结果,验证了本文所提出的算法在中低信噪比情况下的分类性能要优于其他传统算法.A new method to voiced/unvoiced/silence of speech classification using Support Vector Machine （SVM） is proposed. This classifier can effectively classify speech frames into voiced frame, unvoiced frame and silence frame under various levels of signal noise ratio. Firstly,in high SNR, the VU/S classification is done by using the four difference characteristic parameters used in G. 729B VAD as SVM＇s input features. The comparison of experiment resuits shows that the proposed method outperforms other traditional methods （G. 729B VAD and BP network）, which shows the SVM＇s classification method is available. And the performance of SVM for different kernel functions in the experiment was analyzed and discussed as well. Secondly, the paper also discusses the extraction of the statistical features which is immune to the background noise and the adaptive estimation method for the time-varying background noise in low SNR, which are analyzed by applying a statistical model. Lastly, the comparison experiment results in various noise environments under varying levels of SNR are given. According to the simulation results, the proposed method shows significantly better classification performances than the other traditional methods in middle and low SNR cases.

关 键 词：支持向量机 统计学习 统计信号处理 模式识别 语音编码 

分 类 号：TN912.3[电子电信—通信与信息系统]